harn_skills/
lib.rs

1//! Embedded Harn skill corpus.
2//!
3//! This crate exposes the bundled corpus as metadata plus `SKILL.md`
4//! bodies. CLI commands that enumerate, dump, or install these skills
5//! are layered above this foundation.
6
7use std::collections::BTreeMap;
8use std::env;
9use std::fmt;
10use std::fs;
11use std::io;
12use std::path::{Path, PathBuf};
13use std::sync::OnceLock;
14
15/// Environment override for canonical Harn skill discovery.
16pub const HARN_SKILLS_DIR_ENV: &str = "HARN_SKILLS_DIR";
17
18/// Frontmatter fields embedded with each bundled skill.
19#[derive(Debug, Clone, Copy, PartialEq, Eq)]
20pub struct SkillFrontmatter {
21    pub name: &'static str,
22    pub short: &'static str,
23    pub description: &'static str,
24    pub when_to_use: Option<&'static str>,
25}
26
27/// A single skill embedded into the Harn build.
28#[derive(Debug, Clone, Copy, PartialEq, Eq)]
29pub struct EmbeddedSkill {
30    pub name: &'static str,
31    pub frontmatter: SkillFrontmatter,
32    pub body: &'static str,
33    /// The full original SKILL.md source — frontmatter delimiter,
34    /// frontmatter block, blank line, and body — exactly as embedded.
35    /// Use this when round-tripping a skill back to disk so the dumped
36    /// copy is byte-identical to the binary's canonical record.
37    pub source: &'static str,
38}
39
40/// Owned frontmatter fields loaded from a `SKILL.md` on disk.
41#[derive(Debug, Clone, PartialEq, Eq)]
42pub struct DiskSkillFrontmatter {
43    pub name: String,
44    pub short: String,
45    pub description: String,
46    pub when_to_use: Option<String>,
47}
48
49/// A single skill discovered recursively from `HARN_SKILLS_DIR`.
50#[derive(Debug, Clone, PartialEq, Eq)]
51pub struct DiskSkill {
52    pub name: String,
53    pub frontmatter: DiskSkillFrontmatter,
54    pub body: String,
55    pub source: String,
56    pub path: PathBuf,
57}
58
59/// The active canonical corpus used by `harn skill list/get`.
60#[derive(Debug, Clone, PartialEq, Eq)]
61pub enum SkillCorpus {
62    Embedded(&'static [EmbeddedSkill]),
63    Disk(Vec<DiskSkill>),
64}
65
66impl SkillCorpus {
67    pub fn is_disk(&self) -> bool {
68        matches!(self, Self::Disk(_))
69    }
70
71    pub fn len(&self) -> usize {
72        match self {
73            Self::Embedded(skills) => skills.len(),
74            Self::Disk(skills) => skills.len(),
75        }
76    }
77
78    pub fn is_empty(&self) -> bool {
79        self.len() == 0
80    }
81}
82
83/// Error returned when disk skill discovery finds malformed files.
84#[derive(Debug)]
85pub enum SkillDiscoveryError {
86    Io {
87        path: PathBuf,
88        source: io::Error,
89    },
90    MissingFrontmatter {
91        path: PathBuf,
92    },
93    MissingField {
94        path: PathBuf,
95        field: &'static str,
96    },
97    DuplicateName {
98        name: String,
99        first: PathBuf,
100        second: PathBuf,
101    },
102}
103
104impl fmt::Display for SkillDiscoveryError {
105    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
106        match self {
107            Self::Io { path, source } => write!(f, "{}: {source}", path.display()),
108            Self::MissingFrontmatter { path } => {
109                write!(f, "{}: missing SKILL.md frontmatter", path.display())
110            }
111            Self::MissingField { path, field } => {
112                write!(f, "{}: missing `{field}` frontmatter field", path.display())
113            }
114            Self::DuplicateName {
115                name,
116                first,
117                second,
118            } => write!(
119                f,
120                "duplicate skill `{name}` in {} and {}",
121                first.display(),
122                second.display()
123            ),
124        }
125    }
126}
127
128impl std::error::Error for SkillDiscoveryError {}
129
130const SOURCES: &[&str] = &[
131    include_str!("corpus/harn-agent/SKILL.md"),
132    include_str!("corpus/harn-apps/SKILL.md"),
133    include_str!("corpus/harn-de-slop/SKILL.md"),
134    include_str!("corpus/harn-diagnostics/SKILL.md"),
135    include_str!("corpus/harn-docs/SKILL.md"),
136    include_str!("corpus/harn-language/SKILL.md"),
137    include_str!("corpus/harn-orchestration/SKILL.md"),
138    include_str!("corpus/harn-probe/SKILL.md"),
139    include_str!("corpus/harn-product-quality/SKILL.md"),
140    include_str!("corpus/harn-providers/SKILL.md"),
141    include_str!("corpus/harn-rules/SKILL.md"),
142    include_str!("corpus/harn-testing/SKILL.md"),
143    include_str!("corpus/harn-tracing/SKILL.md"),
144    include_str!("corpus/release-harn/SKILL.md"),
145];
146
147static EMBEDDED_SKILLS: OnceLock<Box<[EmbeddedSkill]>> = OnceLock::new();
148
149/// Return every skill bundled into this build.
150pub fn list_embedded_skills() -> &'static [EmbeddedSkill] {
151    EMBEDDED_SKILLS
152        .get_or_init(|| SOURCES.iter().map(|source| parse_skill(source)).collect())
153        .as_ref()
154}
155
156/// Return one bundled skill by canonical skill name.
157pub fn get_embedded_skill(name: &str) -> Option<&'static EmbeddedSkill> {
158    list_embedded_skills()
159        .iter()
160        .find(|skill| skill.name == name)
161}
162
163/// Return the active canonical corpus. `HARN_SKILLS_DIR` wins only
164/// when it contains at least one recursively discovered `SKILL.md`;
165/// otherwise callers fall back to the embedded corpus.
166pub fn resolve_skill_corpus_from_env() -> Result<SkillCorpus, SkillDiscoveryError> {
167    let Ok(dir) = env::var(HARN_SKILLS_DIR_ENV) else {
168        return Ok(SkillCorpus::Embedded(list_embedded_skills()));
169    };
170    if dir.trim().is_empty() {
171        return Ok(SkillCorpus::Embedded(list_embedded_skills()));
172    }
173
174    let skills = list_disk_skills(dir)?;
175    if skills.is_empty() {
176        Ok(SkillCorpus::Embedded(list_embedded_skills()))
177    } else {
178        Ok(SkillCorpus::Disk(skills))
179    }
180}
181
182/// Recursively discover `SKILL.md` files under `root`.
183///
184/// A missing root is treated as an empty disk corpus so
185/// `HARN_SKILLS_DIR` can fall back cleanly to embedded skills.
186pub fn list_disk_skills(root: impl AsRef<Path>) -> Result<Vec<DiskSkill>, SkillDiscoveryError> {
187    let root = root.as_ref();
188    if !root.exists() {
189        return Ok(Vec::new());
190    }
191
192    let mut paths = Vec::new();
193    collect_skill_paths(root, &mut paths)?;
194    paths.sort();
195
196    let mut by_name: BTreeMap<String, DiskSkill> = BTreeMap::new();
197    for path in paths {
198        let skill = parse_disk_skill(&path)?;
199        if let Some(first) = by_name.get(&skill.name) {
200            return Err(SkillDiscoveryError::DuplicateName {
201                name: skill.name,
202                first: first.path.clone(),
203                second: path,
204            });
205        }
206        by_name.insert(skill.name.clone(), skill);
207    }
208
209    Ok(by_name.into_values().collect())
210}
211
212fn parse_skill(source: &'static str) -> EmbeddedSkill {
213    let (frontmatter, body) = split_frontmatter(source);
214    let frontmatter = parse_frontmatter(frontmatter);
215    EmbeddedSkill {
216        name: frontmatter.name,
217        frontmatter,
218        body,
219        source,
220    }
221}
222
223fn collect_skill_paths(dir: &Path, out: &mut Vec<PathBuf>) -> Result<(), SkillDiscoveryError> {
224    let entries = fs::read_dir(dir).map_err(|source| SkillDiscoveryError::Io {
225        path: dir.to_path_buf(),
226        source,
227    })?;
228    for entry in entries {
229        let entry = entry.map_err(|source| SkillDiscoveryError::Io {
230            path: dir.to_path_buf(),
231            source,
232        })?;
233        let path = entry.path();
234        let file_type = entry
235            .file_type()
236            .map_err(|source| SkillDiscoveryError::Io {
237                path: path.clone(),
238                source,
239            })?;
240        if file_type.is_dir() {
241            collect_skill_paths(&path, out)?;
242        } else if file_type.is_file() && entry.file_name() == "SKILL.md" {
243            out.push(path);
244        }
245    }
246    Ok(())
247}
248
249fn parse_disk_skill(path: &Path) -> Result<DiskSkill, SkillDiscoveryError> {
250    let source = fs::read_to_string(path).map_err(|source| SkillDiscoveryError::Io {
251        path: path.to_path_buf(),
252        source,
253    })?;
254    let (frontmatter, body) =
255        split_disk_frontmatter(&source).ok_or_else(|| SkillDiscoveryError::MissingFrontmatter {
256            path: path.to_path_buf(),
257        })?;
258    let frontmatter = parse_disk_frontmatter(path, frontmatter)?;
259    Ok(DiskSkill {
260        name: frontmatter.name.clone(),
261        frontmatter,
262        body: body.to_string(),
263        source,
264        path: path.to_path_buf(),
265    })
266}
267
268fn split_disk_frontmatter(source: &str) -> Option<(&str, &str)> {
269    split_frontmatter_parts(source)
270}
271
272fn parse_disk_frontmatter(
273    path: &Path,
274    frontmatter: &str,
275) -> Result<DiskSkillFrontmatter, SkillDiscoveryError> {
276    let mut name = None;
277    let mut short = None;
278    let mut description = None;
279    let mut when_to_use = None;
280
281    for line in frontmatter.lines() {
282        let Some((key, value)) = line.split_once(':') else {
283            continue;
284        };
285        let value = value.trim().to_string();
286        match key {
287            "name" => name = Some(value),
288            "short" => short = Some(value),
289            "description" => description = Some(value),
290            "when_to_use" => when_to_use = Some(value),
291            _ => {}
292        }
293    }
294
295    Ok(DiskSkillFrontmatter {
296        name: require_disk_field(path, name, "name")?,
297        short: short.unwrap_or_default(),
298        description: require_disk_field(path, description, "description")?,
299        when_to_use,
300    })
301}
302
303fn require_disk_field(
304    path: &Path,
305    value: Option<String>,
306    field: &'static str,
307) -> Result<String, SkillDiscoveryError> {
308    value.ok_or_else(|| SkillDiscoveryError::MissingField {
309        path: path.to_path_buf(),
310        field,
311    })
312}
313
314fn split_frontmatter(source: &'static str) -> (&'static str, &'static str) {
315    let Some((after_open, line_ending)) = split_opening_frontmatter(source) else {
316        panic!("embedded skill source is missing opening frontmatter delimiter");
317    };
318    let Some((frontmatter, body)) = split_closing_frontmatter(after_open, line_ending) else {
319        panic!("embedded skill source is missing closing frontmatter delimiter");
320    };
321    (frontmatter, body)
322}
323
324fn split_frontmatter_parts(source: &str) -> Option<(&str, &str)> {
325    let (after_open, line_ending) = split_opening_frontmatter(source)?;
326    split_closing_frontmatter(after_open, line_ending)
327}
328
329fn split_opening_frontmatter(source: &str) -> Option<(&str, &str)> {
330    if let Some(after_open) = source.strip_prefix("---\n") {
331        Some((after_open, "\n"))
332    } else if let Some(after_open) = source.strip_prefix("---\r\n") {
333        Some((after_open, "\r\n"))
334    } else {
335        None
336    }
337}
338
339fn split_closing_frontmatter<'a>(
340    after_open: &'a str,
341    line_ending: &str,
342) -> Option<(&'a str, &'a str)> {
343    let close = format!("{line_ending}---{line_ending}");
344    let close_offset = after_open.find(&close)?;
345    Some((
346        &after_open[..close_offset],
347        &after_open[close_offset + close.len()..],
348    ))
349}
350
351fn parse_frontmatter(frontmatter: &'static str) -> SkillFrontmatter {
352    let mut name = None;
353    let mut short = None;
354    let mut description = None;
355    let mut when_to_use = None;
356
357    for line in frontmatter.lines() {
358        let Some((key, value)) = line.split_once(':') else {
359            continue;
360        };
361        let value = value.trim();
362        match key {
363            "name" => name = Some(value),
364            "short" => short = Some(value),
365            "description" => description = Some(value),
366            "when_to_use" => when_to_use = Some(value),
367            _ => {}
368        }
369    }
370
371    SkillFrontmatter {
372        name: name.expect("embedded skill frontmatter is missing `name`"),
373        short: short.expect("embedded skill frontmatter is missing `short`"),
374        description: description.expect("embedded skill frontmatter is missing `description`"),
375        when_to_use,
376    }
377}
378
379#[cfg(test)]
380mod tests {
381    use super::*;
382    use std::collections::BTreeSet;
383    use tempfile::TempDir;
384
385    #[test]
386    fn lists_expected_initial_corpus() {
387        let skills = list_embedded_skills();
388        let names: Vec<&str> = skills.iter().map(|skill| skill.name).collect();
389        assert_eq!(
390            names,
391            [
392                "harn-agent",
393                "harn-apps",
394                "harn-de-slop",
395                "harn-diagnostics",
396                "harn-docs",
397                "harn-language",
398                "harn-orchestration",
399                "harn-probe",
400                "harn-product-quality",
401                "harn-providers",
402                "harn-rules",
403                "harn-testing",
404                "harn-tracing",
405                "release-harn",
406            ]
407        );
408        assert_eq!(skills.len(), SOURCES.len());
409    }
410
411    #[test]
412    fn can_fetch_harn_language_skill() {
413        let skill = get_embedded_skill("harn-language").expect("harn-language skill is embedded");
414        assert_eq!(skill.frontmatter.name, "harn-language");
415        assert!(skill.body.contains("Harn language"));
416    }
417
418    #[test]
419    fn skills_have_unique_names_and_body_only_content() {
420        let mut names = BTreeSet::new();
421        for skill in list_embedded_skills() {
422            assert_eq!(skill.name, skill.frontmatter.name);
423            assert!(names.insert(skill.name), "duplicate skill {}", skill.name);
424            assert!(
425                !skill.body.trim().is_empty(),
426                "{} body is empty",
427                skill.name
428            );
429            assert!(
430                !skill.body.trim_start().starts_with("---"),
431                "{} body includes frontmatter",
432                skill.name
433            );
434        }
435    }
436
437    #[test]
438    fn skills_are_sorted_by_name() {
439        let names: Vec<&str> = list_embedded_skills()
440            .iter()
441            .map(|skill| skill.name)
442            .collect();
443        let mut sorted = names.clone();
444        sorted.sort_unstable();
445        assert_eq!(names, sorted);
446    }
447
448    #[test]
449    fn source_round_trips_to_frontmatter_and_body() {
450        for skill in list_embedded_skills() {
451            assert!(
452                split_frontmatter_parts(skill.source).is_some(),
453                "{} source missing opening fence",
454                skill.name
455            );
456            assert!(
457                skill.source.ends_with(skill.body),
458                "{} source must end with the body so dump output is byte-stable",
459                skill.name
460            );
461            assert!(
462                skill.source.contains(&format!("name: {}\n", skill.name))
463                    || skill.source.contains(&format!("name: {}\r\n", skill.name)),
464                "{} source missing canonical name field",
465                skill.name
466            );
467        }
468    }
469
470    #[test]
471    fn frontmatter_split_accepts_crlf_sources() {
472        let source = "---\r\nname: crlf\r\n---\r\n# Body\r\n";
473        let (frontmatter, body) = split_frontmatter_parts(source).expect("CRLF frontmatter");
474        assert_eq!(frontmatter, "name: crlf");
475        assert_eq!(body, "# Body\r\n");
476    }
477
478    #[test]
479    fn frontmatter_split_rejects_missing_closing_fence() {
480        assert!(split_frontmatter_parts("---\nname: missing\n# Body\n").is_none());
481    }
482
483    #[test]
484    fn embedded_corpus_stays_within_binary_budget() {
485        let bytes: usize = SOURCES.iter().map(|source| source.len()).sum();
486        assert!(
487            bytes <= 200 * 1024,
488            "embedded corpus is {bytes} bytes, expected <= 200 KiB"
489        );
490    }
491
492    #[test]
493    fn skill_bodies_are_focused_and_not_placeholders() {
494        let expectations = [
495            ("harn-agent", ["agent_loop", "session id", "approval"]),
496            ("harn-apps", ["std/ui", "ui.test.run", "harn app run"]),
497            ("harn-de-slop", ["typed", "structural guard", "rebase"]),
498            ("harn-diagnostics", ["diagnostic", "repair", "conformance"]),
499            ("harn-docs", ["diátaxis", "glossary", "slopwash"]),
500            ("harn-language", ["quickref", "type", "conformance"]),
501            ("harn-orchestration", ["agent_loop", "workflow", "host"]),
502            ("harn-probe", ["probe", "fact", "evidence"]),
503            (
504                "harn-product-quality",
505                ["canonical path", "control", "liveness"],
506            ),
507            ("harn-providers", ["llm_call", "provider", "schema"]),
508            (
509                "harn-testing",
510                ["conformance", "deterministic", "mock_time"],
511            ),
512            ("harn-tracing", ["replay", "receipts", "transcript"]),
513            ("release-harn", ["release_ship", "merge queue", "tag"]),
514        ];
515
516        for (name, terms) in expectations {
517            let skill = get_embedded_skill(name).expect("expected embedded skill");
518            let body = skill.body.to_ascii_lowercase();
519            assert!(
520                !body.contains("embedded stub") && !body.contains("placeholder"),
521                "{name} should contain real guidance, not stub wording"
522            );
523            for term in terms {
524                assert!(
525                    body.contains(term),
526                    "{name} body should mention focused term `{term}`"
527                );
528            }
529        }
530    }
531
532    #[test]
533    fn skill_bodies_match_split_skill_contract() {
534        for skill in list_embedded_skills() {
535            let lines = skill.body.lines().count();
536            assert!(
537                lines >= 80,
538                "{} body is {lines} lines, expected at least 80",
539                skill.name
540            );
541            assert!(
542                lines <= 300,
543                "{} body is {lines} lines, expected at most 300",
544                skill.name
545            );
546        }
547    }
548
549    #[test]
550    fn language_skill_names_authoritative_spec_sources() {
551        let skill = get_embedded_skill("harn-language").expect("language skill");
552        assert!(
553            skill
554                .body
555                .contains("authoritative chapters under `spec/chapters/*.md`"),
556            "harn-language should direct spec edits to the registered source files"
557        );
558        for generated_edit_target in [
559            "Edit `spec/HARN_SPEC.md`",
560            "Edit `docs/src/language-spec.md`",
561        ] {
562            assert!(
563                !skill.body.contains(generated_edit_target),
564                "harn-language must not direct edits to generated projection `{generated_edit_target}`"
565            );
566        }
567    }
568
569    #[test]
570    fn skill_cross_links_resolve_to_embedded_skills() {
571        let names: BTreeSet<&str> = list_embedded_skills()
572            .iter()
573            .map(|skill| skill.name)
574            .collect();
575        for skill in list_embedded_skills() {
576            for reference in bracketed_skill_references(skill.body) {
577                assert!(
578                    names.contains(reference),
579                    "{} links to unknown embedded skill [[{}]]",
580                    skill.name,
581                    reference
582                );
583            }
584        }
585    }
586
587    #[test]
588    fn diagnostics_skill_mentions_all_code_categories() {
589        let skill = get_embedded_skill("harn-diagnostics").expect("diagnostics skill");
590        for category in [
591            "TYP", "PAR", "NAM", "CAP", "LLM", "ORC", "STD", "PRM", "MOD", "LNT", "FMT", "IMP",
592            "OWN", "RCV", "MAT",
593        ] {
594            assert!(
595                skill.body.contains(&format!("`{category}`")),
596                "harn-diagnostics should mention diagnostic category `{category}`"
597            );
598        }
599    }
600
601    #[test]
602    fn disk_discovery_finds_recursive_skill_files_sorted_by_name() {
603        let temp = TempDir::new().expect("temp dir");
604        write_skill(
605            &temp.path().join("zeta").join("SKILL.md"),
606            "zeta-skill",
607            "Zeta",
608        );
609        write_skill(
610            &temp.path().join("nested").join("alpha").join("SKILL.md"),
611            "alpha-skill",
612            "Alpha",
613        );
614
615        let skills = list_disk_skills(temp.path()).expect("discover disk skills");
616        let names: Vec<&str> = skills.iter().map(|skill| skill.name.as_str()).collect();
617        assert_eq!(names, ["alpha-skill", "zeta-skill"]);
618        assert_eq!(skills[0].frontmatter.description, "Alpha description");
619        assert!(skills[0].body.contains("Alpha body"));
620    }
621
622    #[test]
623    fn disk_discovery_treats_missing_root_as_empty() {
624        let temp = TempDir::new().expect("temp dir");
625        let skills = list_disk_skills(temp.path().join("missing")).expect("discover disk skills");
626        assert!(skills.is_empty());
627    }
628
629    #[test]
630    fn disk_discovery_rejects_duplicate_skill_names() {
631        let temp = TempDir::new().expect("temp dir");
632        write_skill(
633            &temp.path().join("one").join("SKILL.md"),
634            "same-skill",
635            "One",
636        );
637        write_skill(
638            &temp.path().join("two").join("SKILL.md"),
639            "same-skill",
640            "Two",
641        );
642
643        let error = list_disk_skills(temp.path()).expect_err("duplicate name should fail");
644        assert!(
645            error.to_string().contains("duplicate skill `same-skill`"),
646            "unexpected error: {error}"
647        );
648    }
649
650    fn write_skill(path: &Path, name: &str, label: &str) {
651        fs::create_dir_all(path.parent().expect("skill parent")).expect("create skill parent");
652        fs::write(
653            path,
654            format!(
655                "---\nname: {name}\nshort: {label} short\ndescription: {label} description\n---\n# {label}\n\n{label} body\n"
656            ),
657        )
658        .expect("write SKILL.md");
659    }
660
661    fn bracketed_skill_references(body: &str) -> Vec<&str> {
662        let mut references = Vec::new();
663        let mut rest = body;
664        while let Some(start) = rest.find("[[") {
665            rest = &rest[start + 2..];
666            let Some(end) = rest.find("]]") else {
667                break;
668            };
669            references.push(&rest[..end]);
670            rest = &rest[end + 2..];
671        }
672        references
673    }
674}
harn_skills/lib.rs

harn_skills/
lib.rs
