fold/docs/internal/fold-skill.md

2.5 KiB

Fold skill maintenance

The authoritative agent skill is skills/fold-sast/SKILL.md. Do not edit a second prose copy in this directory.

scripts/generate_embedded_skill.py converts the canonical Markdown into the C++ header used by fold skill. Meson generates the build header whenever the canonical skill or generator changes. The checked-in include/fold/cli/embedded_skill.hpp is a reviewable generated mirror and must remain byte-equivalent to the same generator output.

Regenerate and verify the mirror with:

python3 scripts/generate_embedded_skill.py \
  --input skills/fold-sast/SKILL.md \
  --output include/fold/cli/embedded_skill.hpp
python3 scripts/generate_embedded_skill.py \
  --input skills/fold-sast/SKILL.md \
  --output include/fold/cli/embedded_skill.hpp --check

After building, verify the user-facing artifact exactly matches its source and that version provenance agrees across operator artifacts:

build/release/src/cli/fold skill -o /tmp/fold-skill.md
cmp skills/fold-sast/SKILL.md /tmp/fold-skill.md
build/release/src/cli/fold --version
# Compare that value with SARIF runs[0].tool.driver.version,
# native JSON tool.version, and profile JSON version.

Last workflow validation

On 2026-07-29 the canonical recipes were exercised before adoption:

Workflow Evidence Result
Source Clone scan over the exact.cpp and renamed.cpp fixtures Exit 1; expected actionable finding and valid SARIF
Binary Security scan over bof_x64, capped at four functions Exit 0; valid SARIF
Hybrid Merge-only source clone plus binary security scan Exit 1; expected actionable finding and merged SARIF
Query Fixed-string rule query over the source SARIF Exit 1; one deterministic actionable match
Queued MCP fold.scan_start through status, job query, activation, and context fold:mcp_stdio passed
Embedded output fold skill byte comparison with canonical Markdown Passed
Version/artifacts CLI, SARIF, native JSON, profile, and run-artifact provenance One Meson semantic version plus Git revision/dirty state

Exit 1 is the documented success status when actionable findings or query matches exist. No recipe, parser, policy, or tool failure was observed. Full reports and logs were retained under /tmp/fold-task30-examples for that run. No Sage publication or activation was performed.

Do not publish or activate this skill through Sage without explicit approval.