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

harn_skills/
lib.rs
